Add standardized eye-tracking benchmark evaluation module and API end…

[Sahil-aka]

Summary

This PR introduces a standardized benchmarking module for evaluating eye-tracking system performance.

"It provides structured accuracy, precision, and data quality metrics aligned with eye-tracking validation and benchmarking best practices."

Motivation

Currently, the project does not provide a reproducible and standardized way to evaluate eye-tracking accuracy and precision across sessions or devices.

This implementation enables:

Consistent benchmark evaluation

Scientific RMS-based precision measurement

Per-target accuracy analysis

Reproducible performance comparison across setups

Key Additions

Accuracy metrics (mean, median, p95) in pixels and degrees

Scientifically grounded RMS precision metric

Data quality reporting (sample count, loss %)

Per-target accuracy breakdown

New API endpoint:

POST /api/session/benchmark

Input validation and NumPy-safe JSON serialization

Expected Behavior

POST /api/session/benchmark

The API returns:

Overall benchmark metrics

Per-target accuracy breakdown

Precision metrics

Data quality metrics

All returned values are JSON-serializable and numerically stable.

[Sahil-aka]

Thanks for the earlier feedback.

This update includes:

• Refactored benchmark endpoint to be session-based (/api/session/<session_id>/benchmark)
• Metrics now use session metadata instead of request parameters
• Added minimum sample validation warning for small datasets
• Added optional benchmark PDF report generation endpoint

Happy to adjust further if this should align differently with the project architecture.